An effective algorithm for multiagent dispatching of resources in heterogeneous cloud environments
Anatoly Kalyaev, Igor Kalyaev, Maxim Khisamutdinov, Iakov Korovin, Gerald Schaefer · 2016
In this paper, we deal with the problem of adaptive dispatching (distribution) of resources in a cloud computer environment (CCE). In particular, the CCE comprises heterogeneous computational resources and we consider the implementation of a flow of large-scale scientific tasks consisting of a set of information-dependent subtasks that enter the CCE at random times. We present an effective method for multiagent dispatching of CCE resources that can be employed when CCE resources have different performances for different tasks and different channel bandwidths for connecting with cloud infrastructure. We give a formal definition of the problem of CCE dispatching for these conditions and suggest principles of its solution using a set of program agents that are physically implemented on separate CCE resources and that represent their “interests” during dispatch. We introduce the developed algorithm of a program agent as a multiagent dispatcher providing adaptive distribution of all currently free CCE computational resources among the tasks of the flow, taking into consideration their real performances and channel capacities. We further present experimental results obtained using a program model of a heterogeneous CCE.